Phil Yang [Wed, 12 Sep 2018 01:54:26 +0000 (09:54 +0800)]
app/testpmd: optimize mbuf pool allocation
By default, testpmd will create membuf pool for all NUMA nodes and
ignore EAL configuration.
Count the number of available NUMA according to EAL core mask or core
list configuration. Optimized by only creating membuf pool for those
nodes.
Fixes: c9cafcc82de8 ("app/testpmd: fix mempool creation by socket id") Cc: stable@dpdk.org Signed-off-by: Phil Yang <phil.yang@arm.com> Acked-by: Gavin Hu <gavin.hu@arm.com> Acked-by: Bernard Iremonger <bernard.iremonger@intel.com>
mem: support using memfd segments for in-memory mode
Enable using memfd-created segments if supported by the system.
This will allow having real fd's for pages but without hugetlbfs
mounts, which will enable in-memory mode to be used with virtio.
The implementation is mostly piggy-backing on existing real-fd
code, except that we no longer need to unlink any files or track
per-page locks in single-file segments mode, because in-memory
mode does not support secondary processes anyway.
We move some checks from EAL command-line parsing code to memalloc
because it is now possible to use single-file segments mode with
in-memory mode, but only if memfd is supported.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
In a few cases, user may need to query offset into fd for a
particular memory segment (for example, to selectively map
pages). This commit adds a new API to do that.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Now that we can retrieve page fd's internally, we can expose it
as an external API. This will add two flavors of API - thread-safe
and non-thread-safe. Fix up internal API's to return values we need
without modifying rte_errno internally if called from within EAL.
We do not want calling code to accidentally close an internal fd, so
we make a duplicate of it before we return it to the user. Caller is
therefore responsible for closing this fd.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Enable setting and retrieving segment fd's internally.
For now, retrieving fd's will not be used anywhere until we
get an external API, but it will be useful for things like
virtio, where we wish to share segment fd's.
Setting segment fd's will not be available as a public API
at this time, but internally it is needed for legacy mode,
because we're not allocating our hugepages in memalloc in
legacy mode case, and we still need to store the fd.
Another user of get segment fd API is memseg info dump, to
show which pages use which fd's.
Not supported on FreeBSD.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Previously, we were only tracking lock file fd's in single-file
segments mode, but did not track fd's in non-single file mode
because we didn't need to (mmap() call still kept the lock). Now
that we are going to expose these fd's to the world, we need to
have access to them, so track them even in non-single file
segments mode.
We don't need to close fd's after mmap() because we're still
tracking them in an fd list. Also, for anonymous hugepages mode,
fd will always be -1 so exit early on error.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Previously, we were only using lock lists to store per-page lock fd's
because we cannot use modern fcntl() file description locks to lock
parts of the page in single file segments mode.
Now, we will be using this list to store either lock fd's (along with
memseg list fd) in single file segments mode, or per-page fd's (and set
memseg list fd to -1), so rename the list accordingly.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Previously, when we allocated hugepages, we closed the fd's corresponding
to them after we've done our mappings. Since we did mmap(), we didn't
actually lose the reference, but file descriptors used for mmap() do not
count against the fd limit. Since we are going to store all of our fd's,
we will hit the fd limit much more often when using smaller page sizes.
Fix this to raise the fd limit to maximum unconditionally.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
In noshconf mode, no shared files are created, but we're still trying
to unlink them, resulting in detach/destroy failure even though it
should have succeeded. Fix it by exiting early in noshconf mode.
Fixes: 3ee2cde248a7 ("fbarray: support --no-shconf mode") Cc: stable@dpdk.org Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The strncpy function has long been deemed unsafe for use,
in favor of strlcpy or snprintf.
While snprintf is standard and strlcpy is still largely available,
they both have issues regarding error checking and performance.
Both will force reading the source buffer past the requested size
if the input is not a proper c-string, and will return the expected
number of bytes copied, meaning that error checking needs to verify
that the number of bytes copied is not superior to the destination
size.
This contributes to awkward code flow, unclear error checking and
potential issues with malformed input.
The function strscpy has been discussed for some time already and
has been made available in the linux kernel[1].
David Marchand [Mon, 17 Sep 2018 12:45:09 +0000 (14:45 +0200)]
mbuf: remove deprecated segment free functions
__rte_mbuf_raw_free and __rte_pktmbuf_prefree_seg have been deprecated for
a long time now (early 17.05), are not part of the abi and are easily
replaced with existing api.
Signed-off-by: David Marchand <david.marchand@6wind.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Note that the library built by meson will not have the _uio suffix:
librte_pmd_vmxnet3.so - as it follows the directory name, while the
legacy makefile rename it to librte_pmd_vmxnet3_uio.so.
Bruce Richardson [Tue, 18 Sep 2018 10:55:52 +0000 (11:55 +0100)]
build: fix compatibility with meson 0.41 onwards
Versions of meson prior to 0.47 flattened the parameters to the
"set_variable" function, which meant that the function could not take
array variables as a parameter. Therefore, we need to disable driver
tracking for those older versions, in order to maintain compatibility
with the minimum supported 0.41 version, and also v0.45 shipped in
Ubuntu 18.04 release.
Fixes: 806c45dd483d ("build: add configuration summary at end of config") Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Tested-by: Timothy Redaelli <tredaelli@redhat.com>
Bruce Richardson [Fri, 14 Sep 2018 16:17:17 +0000 (17:17 +0100)]
devtools: use shared libs to save space in build test
For usability, the default build type in meson is static, so that
binaries can be run from the build directory easily. However, static
builds take more space, so for build-testing purposes default to using
shared builds where possible.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Luca Boccassi <bluca@debian.org>
Bruce Richardson [Wed, 29 Aug 2018 16:19:20 +0000 (17:19 +0100)]
build: add configuration summary at end of config
After running meson to configure a DPDK build, it can be useful to know
what was automatically enabled or disabled. Therefore, print out by way of
summary a categorised list of libraries and drivers to be built.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Bruce Richardson [Thu, 19 Jul 2018 14:37:02 +0000 (15:37 +0100)]
build: simplify logic for default library dependencies
EAL is a standard dependency of all libraries, except for those built
before it. We can therefore simplify the logic by just checking if EAL
has been processed, and make it a standard dependency if so.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Install in $kerneldir/../extra/dpdk. Usually $kerneldir should something
like: /lib/modules/$kver/build, so this directory will match the default
one used by legacy makefiles.
Fixes: a52f4574f798 ("igb_uio: build with meson") Cc: stable@dpdk.org Signed-off-by: Luca Boccassi <bluca@debian.org> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Bruce Richardson [Mon, 17 Sep 2018 08:18:00 +0000 (09:18 +0100)]
compat: fix symbol version support with meson
For meson builds, the define to enable the symbol version
macros in rte_compat.h was missing. This led to symbols being
omitted from shared objects. For example, checking rte_distributor.so
with objdump and comparing make and meson built versions:
Integrate accelerated networking support into netvsc PMD.
This allows netvsc to manage VF without using failsafe or vdev_netvsc.
For the exception vswitch path some tests like transmit
get a 22% increase in packets/sec.
For the VF path, the code is slightly shorter but has no
real change in performance.
Pro:
* using netvsc is more like other DPDK NIC's
* the exception packet uses less CPU
* much smaller code size
* no locking required on VF transmit/receive path
* no legacy Linux network device to get mangled by userspace
* much simpler (1K vs 9K) LOC
* unified extended statistics
Con:
* using netvsc has more complex startup model
* no bifurcated driver support
* no flow support (since host does not have flow API).
* no tunnel offload support
* no receive interrupt support
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Implement callback functionality on link state changes.
This is not really driven off of interrupt file descriptor like most other
PMD's. Instead, it happens when a link state change message arrives
in the common ring buffer.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
net/netvsc: exhausting Tx descriptors is not an error
If application sends faster than vswitch can keep up, then the
transmit descriptor pool will be exhausted. This is not a failure
so change the name statistic and don't include it in oerrors.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Removed DEV_RX_OFFLOAD_CRC_STRIP offload flag.
Without any specific Rx offload flag, default behavior by PMDs is to
strip CRC.
PMDs that support keeping CRC should advertise DEV_RX_OFFLOAD_KEEP_CRC
Rx offload capability.
Applications that require keeping CRC should check PMD capability first
and if it is supported can enable this feature by setting
DEV_RX_OFFLOAD_KEEP_CRC in Rx offload flag in rte_eth_dev_configure()
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com> Acked-by: Tomasz Duszynski <tdu@semihalf.com> Acked-by: Shahaf Shuler <shahafs@mellanox.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com> Acked-by: Jan Remes <remes@netcope.com> Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com> Acked-by: Hyong Youb Kim <hyonkim@cisco.com>
Igor Romanov [Wed, 29 Aug 2018 07:51:24 +0000 (08:51 +0100)]
net/bonding: do not ignore RSS key on device config
Bonding driver ignores the value of RSS key (that is set in the port RSS
configuration) in bond_ethdev_configure(). So the only way to set
non-default RSS key is by using rss_hash_update(). This is not an
expected behaviour.
Make the bond_ethdev_configure() set default RSS key only if
requested key is set to NULL.
Fixes: 734ce47f71e0 ("bonding: support RSS dynamic configuration") Cc: stable@dpdk.org Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Chas Williams <chas3@att.com>
Igor Romanov [Wed, 29 Aug 2018 07:48:30 +0000 (08:48 +0100)]
net/bonding: use evenly distributed default RSS RETA
Default Redirection Table that is set in bonding driver is distributed
evenly over all Rx queues only within every RETA group (the first RETA
entries in every group are always start with zero). But in the most
drivers, default RETA is distributed over all Rx queues without sequence
resets in the beginning of a new group, which implies more balanced
per-core load.
Change the default RETA to be evenly distributed over all Rx queues
considering the whole table.
Fixes: 734ce47f71e0 ("bonding: support RSS dynamic configuration") Cc: stable@dpdk.org Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Chas Williams <chas3@att.com>
Shagun Agrawal [Mon, 27 Aug 2018 12:52:32 +0000 (18:22 +0530)]
net/cxgbe: add flow ops to match based on dest MAC
Add flow operations to match packets based on destination MAC address.
Allocate and program hardware MPS table with the destination MAC
address to be matched against. The returned MPS index is then used while
offloading flows to LETCAM (maskfull) and HASH (maskless) filter regions.
Also update existing mac_addr_set() to use the new MPS table API.
Shagun Agrawal [Mon, 27 Aug 2018 12:52:31 +0000 (18:22 +0530)]
net/cxgbe: add API to program hardware MPS table
Add API to program and manage hardware Multi Port Switch table. MPS
holds destination MAC addresses to be matched against incoming packets
for further rule processing. Packets not matching any entry in MPS table
will be dropped by default, unless the underlying port is in promiscuous
mode.
Shagun Agrawal [Mon, 27 Aug 2018 12:52:30 +0000 (18:22 +0530)]
net/cxgbe: add flow operations to offload VLAN actions
Add flow API operations to offload vlan push, pop, and rewrite actions.
For vlan push or rewrite actions, allocate and program an entry from
L2T table. Use the L2T index to program vlan actions for LETCAM
(maskfull) and HASH (maskless) filters.
Shagun Agrawal [Mon, 27 Aug 2018 12:52:29 +0000 (18:22 +0530)]
net/cxgbe: add API to program hardware layer 2 table
Add API to program and manage hardware Layer 2 Table. L2T holds
information necessary to rewrite specific fields in packet, such
as destination MAC address and vlan id.
ethdev: fix missing names in Tx offload name array
Patch 5355f443 added two definitions of DEV_TX_OFFLOAD_xxx.
If new Tx offload capabilities are defined, they also must be mentioned
in rte_tx_offload_names in rte_ethdev.c file.
This patch adds the required lines in array rte_tx_offload_names.
Fixes: 5355f4439e2e ("ethdev: introduce generic IP/UDP tunnel checksum and TSO") Cc: stable@dpdk.org Signed-off-by: Dekel Peled <dekelp@mellanox.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Chas Williams [Mon, 6 Aug 2018 20:05:45 +0000 (16:05 -0400)]
net/i40e: stop LLDP before setting local LLDP MIB
>From the Intel Ethernet Controller X710/XXV710/XL710 Specification
Update:
Starting from NVM 5.02, if the Set Local LLDP MIB command is
received while the DCBx specific agent is stopped, the command
returns an EPERM error. If the command is received while the
LLDP agent is stopped, it sets the local MIB without exchanging
LLDP with peer, and returns SUCCESS.
This results in the harmless, but annoying, diagnostic:
default dcb config fails. err = -53, aq_err = 1.
So, if possible (older firmwares cannot safely stop LLDP), stop the
LLDP daemon when we are in software mod before we attempt to call
i40e_set_dcb_config.
Signed-off-by: Chas Williams <chas3@att.com> Acked-by: Qi Zhang <qi.z.zhang@intel.com>
This patch adds alarm handler, and then i40e
PF will use alarm handler instead of interrupt
handler when device is started and Rx interrupt
mode is disabled. This way will save CPU cycles
during receiving packets.
vhost-user: drop connection on message handling failures
There are a lot of cases where vhost-user massage handling
could fail and end up in a fully not recoverable state. For
example, allocation failures of shadow used ring and batched
copy array are not recoverable and leads to the segmentation
faults like this on the receiving/transmission path:
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7f913fecf0 (LWP 43625)]
in copy_desc_to_mbuf () at /lib/librte_vhost/virtio_net.c:760
760 batch_copy[vq->batch_copy_nb_elems].dst =
This could be easily reproduced in case of low memory or big
number of vhost-user ports.
Fix that by propagating error to the upper layer which will
end up with disconnection in case we can not report to
the message sender when the error happens.
Fixes: f689586bc060 ("vhost: shadow used ring update") Cc: stable@dpdk.org Signed-off-by: Ilya Maximets <i.maximets@samsung.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Eric Zhang [Wed, 29 Aug 2018 15:55:21 +0000 (11:55 -0400)]
net/virtio-user: check negotiated features before set
This patch checks negotiated features to see if necessary to offload
before set the tap device offload capabilities. It also checks if kernel
support the TUNSETOFFLOAD operation.
Ilya Maximets [Wed, 15 Aug 2018 14:54:39 +0000 (17:54 +0300)]
vhost: fix zmbufs array leak after NUMA realloc
'numa_realloc()' allocates 'zmbufs' even if zero copy mode
is not configured. This leads to memory leak, because array
is freed only for zero copy case.
Fixes: 2651726defb7 ("vhost: do deep copy while reallocating queue") CC: stable@dpdk.org Signed-off-by: Ilya Maximets <i.maximets@samsung.com> Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>
Rami Rosen [Fri, 24 Aug 2018 07:43:02 +0000 (10:43 +0300)]
doc: fix wrong usage of bind command
This patch fixes wrong usage of bind command in vhost.rst.
Using "dpdk-devbind.py -b=uio_pci_generic 0000:00:04.0" gives
an error of "unbind failed". It should be "-b uio_pci_generic" so
it will work correctly.
In case of a temporary failure the ixgbe driver can return the internal
error IXGBE_ERR_RESET_FAILED to the application. Instead, return
-EAGAIN as per the public API specification.
Jia Yu [Mon, 20 Aug 2018 05:18:45 +0000 (22:18 -0700)]
net/bonding: fix buffer corruption in packets
When bond slave devices cannot transmit all packets in bufs array,
tx_burst callback shall merge the un-transmitted packets back to
bufs array. Recent merge logic introduced a bug which causes
invalid mbuf addresses being written to bufs array.
When caller frees the un-transmitted packets, due to invalid addresses,
application will crash.
The fix is avoid shifting mbufs, and directly write un-transmitted
packets back to bufs array.
Alejandro Lucero [Fri, 24 Aug 2018 14:25:35 +0000 (15:25 +0100)]
ethdev: fix MAC changes when live change not supported
Current code assumes a MAC change can occur when the port has been
started. In fact, there are some NICs which require this port state
for being successful, but other NICs not always support MAC change
in that case.
This patch supports a new device flag for a device advertising this
limitation, and if the flag is set, the MAC is changed before the
port starts.
Tomasz Duszynski [Fri, 24 Aug 2018 18:30:00 +0000 (20:30 +0200)]
net/mvpp2: fix array initialization
Fix used_bpools array initialization by using range initializer.
This way all necessary variables are properly initialized regardless
of PP2_NUM_PKT_PROC value.
Fixes: 0ddc9b815b11 ("net/mrvl: add net PMD skeleton") Cc: stable@dpdk.org Signed-off-by: Tomasz Duszynski <tdu@semihalf.com>
Liron Himi [Fri, 24 Aug 2018 18:29:58 +0000 (20:29 +0200)]
common/mvep: add common code for Marvell drivers
Add MVEP (Marvell Embedded Processors) to drivers/common which
will keep code reused by current and future MRVL PMDs.
Right now we have only common DMA memory initialization routines there.
Shahaf Shuler [Tue, 14 Aug 2018 11:17:21 +0000 (14:17 +0300)]
net/mlx5: fix RSS flow action hash type selection
On the code after the below commits, the criteria to select the IPV4 or
IPV6 hash functions was the existence of some ETH_RSS_IPV4 RSS types on
the flow rule.
The check is wrong. For example ETH_RSS_NONFRAG_IPV4_TCP will not select
the IPV4 hash which will cause the packet to be spread in a bad way.
Fix it by adding the corresponding types needed for each hash selection.
Shahaf Shuler [Mon, 13 Aug 2018 06:47:57 +0000 (09:47 +0300)]
net/mlx5: disable ConnectX-4 Lx Multi Packet Send by default
On ConnectX-4 Lx the Multi Packet Send (MPW) feature is considered
un-secure, as on some cases were the application provides incorrect mbufs
on the Tx burst the host or NIC can get stuck.
Hence, disabling the feature by default for this specific NIC.
Users can still enable this feature and enjoy the performance gain
(mostly for low number of cores) by using the txq_mpw_en devarg.
This patch will impact the out of the box performance of some application
using ConnectX-4 Lx for the sack of security and robustness.
Since we need different defaults based on the underlying device the mpw
field in the configuration struct was extended to contain also the
MLX5_ARG_UNSET option.
The event buffer was changed to be a fixed size value,
had a couple of issues. The big one is that rte_free was still
being called for a pointer that was not setup with rte_malloc().
The event buffer was also too small to handle heavy receive
traffic; and running the event buffer out would crash
the application.
Fix by going back to a dynamically resized event buffer.
And grow it by 25% to avoid lots of realloc's.
Fixes: 530af95a7849 ("bus/vmbus: avoid signalling host on read") Cc: stable@dpdk.org Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Tune the vmbus connection so the host scans faster. This improves
transmit performance. The host default value is 100us but setting
to 50us reduces packet loss significantly.
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
git.droids-corp.org / dpdk.git / log